Integrating catchment properties in small scale species distribution models of stream macroinvertebrates

Abstract

Species distribution models are increasingly applied to freshwater ecosystems. Most applications use large scales, coarse resolutions and anthropocentric modelling extents, thus not being able to consider important environmental predictors and ecological processes detectable within a catchment and at finer scales. Moreover, high resolution predictions of species distribution in streams can help improve our understanding of how environmental variables within a catchment affect the distribution of stream macroinvertebrates. We built models at a resolution of 25m×25m for a 488km2 catchment in northern Germany to determine whether the spatial approach in which environmental predictors are implemented in the model affects the overall performance. We used predictors from four different categories relevant to freshwater ecosystems: bioclimatic, topographic, hydrologic and land use. Two spatial approaches were tested: a local one, or grid based and a cumulative for the upstream area, or subcatchment specific. Models were evaluated in terms of model performance and accuracy in order to identify the approach best suited for each category, as well as the most important predictor in each. In the case of the land use category, the subcatchment approach made a significant difference, increasing performance. A final model, calibrated with the selected predictors, resulted in the highest model performance and accuracy. Our results indicate that species distribution models perform well and are accurate at high resolutions, within small catchments. We conclude that catchment wide models, especially when using predictors from multiple categories, have the potential to significantly improve modelling framework of species distribution in freshwater ecosystems. The information produced by accurate, small scale, species distribution models can guide managers and conservation practitioners, by predicting the effects of management decisions within a catchment. We suggest that highly resolved predictors be applied in models using the catchment approach.